Grok Bot Shopping + Which AI Models To Pair With It, With Real GoldieBench Scores For Pages, Agents And Grok 4.7

Grok Bot shopping doesn't let you choose its model, so the GoldieBench scores that matter are the ones for the agents and builds you pair with it.

The Grok Bot FAQ says Grok Bot picks the model for each task and there is no model picker.

That changes how you use a leaderboard.

You read it to choose the models that build your deal pages, comparison sheets and order dashboards around the Bot.

If you want my team to help you choose, book a free SEO strategy session.

Why the model choice sits outside Grok Bot

The FAQ says the backend model can be a first-party model or a third-party one, and the choice can differ from one request to the next.

Grok Bot doesn't show the underlying model for a request.

Users can't be routed to or away from a particular model.

So the shopping cards, Buy now flow and tracking added in v0.78.1 on 27 October 2026 run on whatever Grok Bot picks.

Your model decisions apply to everything else in your stack.

That's where GoldieBench helps.

Grok Bot shopping in 60 seconds

Here's what the Bot does, so the model advice has context.

It searches stores and returns product cards with photos, rating, price and store.

Buy now on a product page opens an order that you approve.

After payment, it follows the parcel and reports carrier and stage.

It pays through a wallet in the Vault's Payments tab, such as Shop Pay or X Money, or through a card.

The Vault documentation explains how saved cards, contact info and wallets work.

What GoldieBench actually measures

The methodology page says every model gets the same fixed prompt, one shot, with no iteration and no best-of-N picking.

Every result must be one self-contained HTML file that runs in a browser.

Scores are Julian's own 0 to 10 numbers, and medals go to the top three per task.

That makes the scores easy to check, because every demo on the bench is playable on the site.

None of the tasks is a shopping checkout.

The closest fit is the Page category, which covers the kind of web pages a buying workflow produces.

Grok's own numbers on the bench

On the GoldieBench leaderboard, read on 31 October 2026, Grok sits at #7 with an 8.09 average and 43 of 47 tasks scored.

The Grok benchmark page lists these Page scores:

Two gold medals in three page tasks is a strong page-building record.

Grok 4.7: the reference scores worth knowing

The Grok 4.7 benchmark page shows a 7.15 average across 20 scored game tasks.

It also lists outside benchmarks pulled from xAI and Artificial Analysis.

GDPval is the only Elo-style number in that list, and it's the one closest to everyday business work.

Terminal-Bench 4.0 at 38.0% is the lowest of those outside numbers, and the same deep dive shows its weakest builds at 2.3 and 2.2.

For shopping support work, that points away from long terminal-heavy jobs and towards page and comparison builds.

Which models to pair for shopping pages

The Landing Page task is the best proxy for a product comparison or deal page.

Its page shows a task average of 8.11 across 21 scored models out of 26.

Any of the four 9.0 models is a sound pick for a deal tracker page or a supplier comparison sheet.

The overall leaderboard for agent teams

If you want one model running the rest of your Agent OS next to Grok Bot, look at the overall averages.

These come from the GoldieBench home page leaderboard.

GPT-5.6 Sol is the highest-ranked model on the board whose strongest category is Pages, which suits shopping reports and comparison pages.

What the scores can't tell you about buying

No benchmark on this site measures checkout safety.

For a buying agent, the approval setup matters more than any model score.

Speed, judgement on a product page and care at a checkout aren't scored anywhere here.

Treat a 9.0 as proof a model can build a clean page in one go, not as proof it shops well.

A 7.2 on the same task still produced a working page; the judge just flagged layout problems.

The Grok Bot approvals page says an action that matches an Ask first rule always gets an approval card, and Ask first wins when an Allow rule also matches.

The payment rule that beats any model pick

The release notes and the docs read differently on payments.

xAI's release notes promise you sign off on each purchase (v0.77.0) and that Buy now leads to an order awaiting your approval (v0.78.1).

The Vault docs on paying say a saved card fills in at checkout without asking each time, and no setting makes each use need approval.

Present both to your team and pick the strict option.

Add an Ask first rule for every purchase, and don't save a card in Vault.

Leave Securely save to Bot's vault unchecked when a Secure Form asks for card details.

Or pay through a wallet: the Vault page says a Link one-time card request shows the merchant, the items and the total before you choose Pay.

My pairing setup, by job

• Buying, tracking and subscriptions: Grok Bot itself, with its own model routing and your Ask first rules.
• Deal and comparison pages: one of the Landing Page 9.0 models, such as Grok, Fusion, Fugu Ultra or GLM-5.2.
• Weekly shopping reports: GPT-5.6 Sol, ranked #4 overall and best at Pages.
• The rest of your agent team: Fusion, Claude Opus 5 or Hermes MoA, the top three overall.

Start with one page model, not four.

Build a single deal tracker page, check it against your real product list, and only then add the next model.

Members of the AI Profit Boardroom get the installable Agent OS to run these pairings in one place.

Learn to wire these models into one system inside the Boardroom coaching calls.

Or talk to my team about your setup.

Also On Our Network

• Agent Operatorsthe six-step Vault and Ask first setup I'd run before any checkout
• AI Income Deskthree small business use cases and the purchase policy to set first
• AI Tool Verdictan eight-criteria scorecard of the cards, payments and tracking
• agentos.guidefive copy-paste prompts and a weekly buying workflow for your Agent OS
• aiprofitboardroom.comthe complete Grok Bot Shopping guide on the AI Profit Boardroom blog

Pick your page model from the table above, then book a free strategy session to plan the rest of your Grok Bot shopping stack.